Voice Quality and Between-Frame Entropy for Sleepiness Estimation

Vijay Ravi, Soo Jin Park, Amber Afshan, Abeer Alwan

Sleepiness monitoring and prediction has many potential applications, such as being a safety feature in driver-assistance systems. In this study, we address the ComparE 2019 Continuous Sleepiness task of estimating the degree of sleepiness from voice data. The voice quality feature set was proposed to capture the acoustic characteristics related to the degree of sleepiness of a speaker, and between-frame entropy was proposed as an instantaneous measure of the speaking rate. An outlier elimination on the training data using between-frame entropy enhanced the system robustness in all conditions. This was followed by a regression system to predict the degree of sleepiness. Utterances were represented using i-vectors computed from voice quality features. Similar systems were also developed using mel-frequency cepstral coefficients and the ComParE16 feature set. These three systems were combined using score-level fusion. Results suggested complementarity between these feature sets. The complete system outperformed the baseline system which used the ComParE16 feature set. A relative improvement of 19.5% and 5.4% was achieved on the development and the test datasets, respectively.

 DOI: 10.21437/Interspeech.2019-2988

Cite as: Ravi, V., Park, S.J., Afshan, A., Alwan, A. (2019) Voice Quality and Between-Frame Entropy for Sleepiness Estimation. Proc. Interspeech 2019, 2408-2412, DOI: 10.21437/Interspeech.2019-2988.

  author={Vijay Ravi and Soo Jin Park and Amber Afshan and Abeer Alwan},
  title={{Voice Quality and Between-Frame Entropy for Sleepiness Estimation}},
  booktitle={Proc. Interspeech 2019},